YOLO-Based Missile Pose Estimation Under Uncalibrated Conditions

In missile docking, high-precision section alignment is vital for mission success. Traditional techniques, relying on radar and GPS, face calibration complexities and environmental interference risks, potentially leading to inaccurate estimations. To address this issue, the PoseNoCal-YOLOv5 is developed, offering a vision-based pose estimation approach that requires no traditional calibration, enhancing docking precision and reliability. It comprises two sub-networks: an improved YOLOv5 object detection model with an attention mechanism for precise object detection, and an uncalibrated pose estimation module using re-linearization for pose estimation without camera calibration. A simulated dataset is created for validation, covering diverse docking scenarios. Extensive experiments on this simulated datasets prove the effectiveness of the proposed method.